[420official]

> ... were recognized for creating the tool that has predicted the 3D structures of almost every known protein on the planet.

I wonder if relying on a tool that doesn't 100% accurately represent reality could have a negative effect on future research

[JamesBarney]

Current methods are not 100% accurate either. No study is 100%.

Honestly the only field that has a P value that comes close to 100% is physics. Even medicine which is far more rigorous than most fields fails quite often in phase 3 trials after having vetted it in phase 2.

[tedsanders]

Even physics is nowhere close to “100% accurate.” Most fields of physics approximate many body problems that are infeasible to compute, let alone fully specify. E.g. Astrophysics, solid state physics, nuclear physics, etc. Practitioners regularly use empirically measured parameters like cross-sectional scattering areas, and those parameters are updated and narrowed over time.

[dekhn]

The predictions made by AlphaFold are now indistinguishable from experimental data collection error so folks aren't super concerned. Anyway structures are typically qualititaive tools useful for thinking about proteins, rather than direct targets of computational predictions (hasn't stopped people from trying).

[flobosg]

> that doesn't 100% accurately represent reality

It could be argued that this is the case of every scientific tool ever used.

[summerlight]

If you can develop any kind of non-trivial scientific fields with 100% accuracy and plausible evidence, that finding alone would be qualified for the Nobel prize. Even abstract fields like Math are prone to human errors.

[stainablesteel]

its a decent enough start because scaling the instrumentation-route of doing this is a lot slower than the ML approach, it can only improve